Skip to contentSkip to conversation

Implement use of HW assistance on TLB table walk on 8xx

PATCH Superseded Christophe Leroy 1 participant 16 messages May 29, 2018 – Aug 7, 2018 PowerPC

v3 of 9 Superseded Latest is v9 →

lore
Share & tools
lore
#0 lore raw

v3 cover Implement use of HW assistance on TLB table walk on 8xx

The purpose of this serie is to implement hardware assistance for TLB table walk on the 8xx.

First part is to make L1 entries and L2 entries independant.
For that, we need to alter ioremap functions in order to handle GUARD attribute at the PGD/PMD level.

Last part is to reuse PTE fragment implemented on PPC64 in order to not waste 16k Pages for page tables as only 4k are used.

Tested successfully on 8xx.

Successfull compilation on following defconfigs (v3):
ppc64_defconfig
ppc64e_defconfig

Successfull compilation on following defconfigs (v2):
ppc64_defconfig
ppc64e_defconfig
pseries_defconfig
pmac32_defconfig
linkstation_defconfig
corenet32_smp_defconfig
ppc40x_defconfig
storcenter_defconfig
ppc44x_defconfig

Changes in v3:
- Fixed an issue in the 09/14 when CONFIG_PIN_TLB_TEXT was not enabled
- Added performance measurement in the 09/14 commit log
- Rebased on latest 'powerpc/merge' tree, which conflicted with 13/14

Changes in v2:
- Removed the 3 first patchs which have been applied already
- Fixed compilation errors reported by Michael
- Squashed the commonalisation of ioremap functions into a single patch
- Fixed the use of pte_fragment
- Added a patch optimising perf counting of TLB misses and instructions

Christophe Leroy (14):
Revert "powerpc/8xx: Use L1 entry APG to handle _PAGE_ACCESSED for
CONFIG_SWAP"
powerpc: move io mapping functions into ioremap.c
powerpc: make ioremap_bot common to PPC32 and PPC64
powerpc: common ioremap functions.
powerpc: use _ALIGN_DOWN macro for VMALLOC_BASE powerpc/nohash32: allow setting GUARDED attribute in the PMD directly powerpc/8xx: set GUARDED attribute in the PMD directly powerpc/8xx: Remove PTE_ATOMIC_UPDATES
powerpc/mm: Use hardware assistance in TLB handlers on the 8xx powerpc/8xx: reunify TLB handler routines
powerpc/8xx: Free up SPRN_SPRG_SCRATCH2
powerpc/mm: Make pte_fragment_alloc() common to PPC32 and PPC64 powerpc/mm: Use pte_fragment_alloc() on 8xx
powerpc/8xx: Move SW perf counters in first 32kb of memory

 arch/powerpc/include/asm/book3s/32/pgtable.h |  29 +-
 arch/powerpc/include/asm/book3s/64/pgtable.h |   2 +
 arch/powerpc/include/asm/highmem.h           |  11 -
 arch/powerpc/include/asm/hugetlb.h           |   4 +-
 arch/powerpc/include/asm/machdep.h           |   2 +-
 arch/powerpc/include/asm/mmu-8xx.h           |  38 +--
 arch/powerpc/include/asm/mmu_context.h       |  53 ++++
 arch/powerpc/include/asm/nohash/32/pgalloc.h |  56 +++-
 arch/powerpc/include/asm/nohash/32/pgtable.h |  77 +++--
 arch/powerpc/include/asm/nohash/32/pte-8xx.h |   6 +-
 arch/powerpc/include/asm/nohash/pgtable.h    |   4 +
 arch/powerpc/include/asm/page.h              |   2 +-
 arch/powerpc/include/asm/pgtable-types.h     |   4 +
 arch/powerpc/kernel/head_8xx.S               | 409 +++++++++++----------------
 arch/powerpc/mm/8xx_mmu.c                    |  12 +-
 arch/powerpc/mm/Makefile                     |   2 +-
 arch/powerpc/mm/dma-noncoherent.c            |   2 +-
 arch/powerpc/mm/dump_linuxpagetables.c       |  32 ++-
 arch/powerpc/mm/hugetlbpage.c                |  12 +
 arch/powerpc/mm/init_32.c                    |   6 +-
 arch/powerpc/mm/{pgtable_64.c => ioremap.c}  | 239 ++++++----------
 arch/powerpc/mm/mem.c                        |  16 +-
 arch/powerpc/mm/mmu_context_book3s64.c       |  44 ---
 arch/powerpc/mm/mmu_context_nohash.c         |   4 +
 arch/powerpc/mm/pgtable-book3s64.c           |  72 -----
 arch/powerpc/mm/pgtable.c                    |  82 ++++++
 arch/powerpc/mm/pgtable_32.c                 | 167 +++--------
 arch/powerpc/mm/pgtable_64.c                 | 177 ------------
 arch/powerpc/platforms/Kconfig.cputype       |  19 ++
 29 files changed, 657 insertions(+), 926 deletions(-)
 copy arch/powerpc/mm/{pgtable_64.c => ioremap.c} (53%)
Signature · 1 line
2.13.3
#1 lore raw

PATCH v3 1/14 Revert "powerpc/8xx: Use L1 entry APG to handle _PAGE_ACCESSED for CONFIG_SWAP"

This reverts commit 4f94b2c7462d9720b2afa7e8e8d4c19446bb31ce.

That commit was buggy, as it used rlwinm instead of rlwimi. Instead of fixing that bug, we revert the previous commit in order to reduce the dependency between L1 entries and L2 entries

---
 arch/powerpc/include/asm/mmu-8xx.h | 34 +++++-----------------------
 arch/powerpc/kernel/head_8xx.S     | 45 +++++++++++++++++++++++---------------
 arch/powerpc/mm/8xx_mmu.c          |  2 +-
 3 files changed, 34 insertions(+), 47 deletions(-)
arch/powerpc/include/asm/mmu-8xx.h +6 -28
index 4f547752ae79..193f53116c7a 100644
@@ -34,20 +34,12 @@
  * respectively NA for All or X for Supervisor and no access for User.
  * Then we use the APG to say whether accesses are according to Page rules or
  * "all Supervisor" rules (Access to all)
- * We also use the 2nd APG bit for _PAGE_ACCESSED when having SWAP:
- * When that bit is not set access is done iaw "all user"
- * which means no access iaw page rules.
- * Therefore, we define 4 APG groups. lsb is _PMD_USER, 2nd is _PAGE_ACCESSED
- * 0x => No access => 11 (all accesses performed as user iaw page definition)
- * 10 => No user => 01 (all accesses performed according to page definition)
- * 11 => User => 00 (all accesses performed as supervisor iaw page definition)
+ * Therefore, we define 2 APG groups. lsb is _PMD_USER
+ * 0 => No user => 01 (all accesses performed according to page definition)
+ * 1 => User => 00 (all accesses performed as supervisor iaw page definition)
  * We define all 16 groups so that all other bits of APG can take any value
  */
-#ifdef CONFIG_SWAP
-#define MI_APG_INIT	0xf4f4f4f4
-#else
 #define MI_APG_INIT	0x44444444
-#endif
 
 /* The effective page number register.  When read, contains the information
  * about the last instruction TLB miss.  When MI_RPN is written, bits in
@@ -115,20 +107,12 @@
  * Supervisor and no access for user and NA for ALL.
  * Then we use the APG to say whether accesses are according to Page rules or
  * "all Supervisor" rules (Access to all)
- * We also use the 2nd APG bit for _PAGE_ACCESSED when having SWAP:
- * When that bit is not set access is done iaw "all user"
- * which means no access iaw page rules.
- * Therefore, we define 4 APG groups. lsb is _PMD_USER, 2nd is _PAGE_ACCESSED
- * 0x => No access => 11 (all accesses performed as user iaw page definition)
- * 10 => No user => 01 (all accesses performed according to page definition)
- * 11 => User => 00 (all accesses performed as supervisor iaw page definition)
+ * Therefore, we define 2 APG groups. lsb is _PMD_USER
+ * 0 => No user => 01 (all accesses performed according to page definition)
+ * 1 => User => 00 (all accesses performed as supervisor iaw page definition)
  * We define all 16 groups so that all other bits of APG can take any value
  */
-#ifdef CONFIG_SWAP
-#define MD_APG_INIT	0xf4f4f4f4
-#else
 #define MD_APG_INIT	0x44444444
-#endif
 
 /* The effective page number register.  When read, contains the information
  * about the last instruction TLB miss.  When MD_RPN is written, bits in
@@ -180,12 +164,6 @@
  */
 #define SPRN_M_TW	799
 
-/* APGs */
-#define M_APG0		0x00000000
-#define M_APG1		0x00000020
-#define M_APG2		0x00000040
-#define M_APG3		0x00000060
-
 #ifdef CONFIG_PPC_MM_SLICES
 #include <asm/nohash/32/slice.h>
 #define SLICE_ARRAY_SIZE	(1 << (32 - SLICE_LOW_SHIFT - 1))
arch/powerpc/kernel/head_8xx.S +27 -18

92 lines not on this page · raw message

arch/powerpc/mm/8xx_mmu.c +1 -1
index cf77d755246d..5d53684c2ebd 100644
@@ -79,7 +79,7 @@ void __init MMU_init_hw(void)
 	for (; i < 32 && mem >= LARGE_PAGE_SIZE_8M; i++) {
 		mtspr(SPRN_MD_CTR, ctr | (i << 8));
 		mtspr(SPRN_MD_EPN, (unsigned long)__va(addr) | MD_EVALID);
-		mtspr(SPRN_MD_TWC, MD_PS8MEG | MD_SVALID | M_APG2);
+		mtspr(SPRN_MD_TWC, MD_PS8MEG | MD_SVALID);
 		mtspr(SPRN_MD_RPN, addr | flags | _PAGE_PRESENT);
 		addr += LARGE_PAGE_SIZE_8M;
 		mem -= LARGE_PAGE_SIZE_8M;
Signature · 1 line
2.13.3
#2 lore raw

PATCH v3 2/14 powerpc: move io mapping functions into ioremap.c

In reply to Christophe Leroy

This patch is the first of a serie that intends to make io mappings common to PPC32 and PPC64.

It moves ioremap/unmap fonctions into a new file called ioremap.c with no other modification to the functions.
For the time being, the PPC32 and PPC64 parts get enclosed into #ifdef. Following patches will aim at making those functions as common as possible between PPC32 and PPC64.

This patch also moves EXPORT_SYMBOL at the end of each function

---
 arch/powerpc/mm/Makefile                    |   2 +-
 arch/powerpc/mm/{pgtable_64.c => ioremap.c} | 313 +++++++++++++++-------------
 arch/powerpc/mm/pgtable_32.c                | 139 ------------
 arch/powerpc/mm/pgtable_64.c                | 177 ----------------
 4 files changed, 166 insertions(+), 465 deletions(-)
 copy arch/powerpc/mm/{pgtable_64.c => ioremap.c} (56%)
arch/powerpc/mm/Makefile +1 -1
index f06f3577d8d1..22d54c1d90e1 100644
@@ -9,7 +9,7 @@ ccflags-$(CONFIG_PPC64)	:= $(NO_MINIMAL_TOC)
 
 obj-y				:= fault.o mem.o pgtable.o mmap.o \
 				   init_$(BITS).o pgtable_$(BITS).o \
-				   init-common.o mmu_context.o drmem.o
+				   init-common.o mmu_context.o drmem.o ioremap.o
 obj-$(CONFIG_PPC_MMU_NOHASH)	+= mmu_context_nohash.o tlb_nohash.o \
 				   tlb_nohash_low.o
 obj-$(CONFIG_PPC_BOOK3E)	+= tlb_low_$(BITS)e.o
arch/powerpc/mm/ioremap.c renamed from arch/powerpc/mm/pgtable_64.c +165 -148

396 lines not on this page · raw message

arch/powerpc/mm/pgtable_32.c +0 -139

151 lines not on this page · raw message

arch/powerpc/mm/pgtable_64.c +0 -177

185 lines not on this page · raw message

Signature · 1 line
2.13.3
#3 lore raw

PATCH v3 3/14 powerpc: make ioremap_bot common to PPC32 and PPC64

In reply to Christophe Leroy

Today, early ioremap maps from IOREMAP_BASE down to up on PPC64 and from IOREMAP_TOP up to down on PPC32

This patchs modifies PPC32 behaviour to get same behaviour as PPC64

---
 arch/powerpc/include/asm/book3s/32/pgtable.h | 29 ++++++++++++++++++------
 arch/powerpc/include/asm/highmem.h           | 11 ----------
 arch/powerpc/include/asm/nohash/32/pgtable.h | 33 ++++++++++++++++++----------
 arch/powerpc/mm/dma-noncoherent.c            |  2 +-
 arch/powerpc/mm/dump_linuxpagetables.c       |  6 ++---
 arch/powerpc/mm/init_32.c                    |  6 ++++-
 arch/powerpc/mm/ioremap.c                    | 21 +++++++++---------
 arch/powerpc/mm/mem.c                        |  7 +++---
 8 files changed, 66 insertions(+), 49 deletions(-)
arch/powerpc/include/asm/book3s/32/pgtable.h +22 -7
index c615abdce119..ccf0d77277b1 100644
@@ -50,20 +50,32 @@
  * virtual space that goes below PKMAP and FIXMAP
  */
 #ifdef CONFIG_HIGHMEM
+#ifdef CONFIG_PPC_4K_PAGES
+#define PKMAP_ORDER	PTE_SHIFT
+#else
+#define PKMAP_ORDER	9
+#endif
+#define LAST_PKMAP	(1 << PKMAP_ORDER)
+#ifndef CONFIG_PPC_4K_PAGES
+#define PKMAP_BASE	(FIXADDR_START - PAGE_SIZE*(LAST_PKMAP + 1))
+#else
+#define PKMAP_BASE	((FIXADDR_START - PAGE_SIZE*(LAST_PKMAP + 1)) & PMD_MASK)
+#endif
 #define KVIRT_TOP	PKMAP_BASE
 #else
 #define KVIRT_TOP	(0xfe000000UL)	/* for now, could be FIXMAP_BASE ? */
 #endif
+#define IOREMAP_BASE	VMALLOC_BASE
 
 /*
- * ioremap_bot starts at that address. Early ioremaps move down from there,
- * until mem_init() at which point this becomes the top of the vmalloc
+ * ioremap_bot starts at IOREMAP_BASE. Early ioremaps move up from there,
+ * until mem_init() at which point this becomes the bottom of the vmalloc
  * and ioremap space
  */
 #ifdef CONFIG_NOT_COHERENT_CACHE
-#define IOREMAP_TOP	((KVIRT_TOP - CONFIG_CONSISTENT_SIZE) & PAGE_MASK)
+#define IOREMAP_END	((KVIRT_TOP - CONFIG_CONSISTENT_SIZE) & PAGE_MASK)
 #else
-#define IOREMAP_TOP	KVIRT_TOP
+#define IOREMAP_END	KVIRT_TOP
 #endif
 
 /*
@@ -85,11 +97,12 @@
  */
 #define VMALLOC_OFFSET (0x1000000) /* 16M */
 #ifdef PPC_PIN_SIZE
-#define VMALLOC_START (((_ALIGN((long)high_memory, PPC_PIN_SIZE) + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1)))
+#define VMALLOC_BASE (((_ALIGN((long)high_memory, PPC_PIN_SIZE) + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1)))
 #else
-#define VMALLOC_START ((((long)high_memory + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1)))
+#define VMALLOC_BASE ((((long)high_memory + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1)))
 #endif
-#define VMALLOC_END	ioremap_bot
+#define VMALLOC_START	ioremap_bot
+#define VMALLOC_END	IOREMAP_END
 
 #ifndef __ASSEMBLY__
 #include <linux/sched.h>
@@ -298,6 +311,8 @@ static inline void __ptep_set_access_flags(struct mm_struct *mm,
 
 int map_kernel_page(unsigned long va, phys_addr_t pa, int flags);
 
+#include <asm/fixmap.h>
+
 /* Generic accessors to PTE bits */
 static inline int pte_write(pte_t pte)		{ return !!(pte_val(pte) & _PAGE_RW);}
 static inline int pte_read(pte_t pte)		{ return 1; }
arch/powerpc/include/asm/highmem.h +0 -11
index cec820f961da..2eacea504da2 100644
@@ -44,17 +44,6 @@ extern pte_t *pkmap_page_table;
  * and PKMAP can be placed in a single pte table. We use 512 pages for PKMAP
  * in case of 16K/64K/256K page sizes.
  */
-#ifdef CONFIG_PPC_4K_PAGES
-#define PKMAP_ORDER	PTE_SHIFT
-#else
-#define PKMAP_ORDER	9
-#endif
-#define LAST_PKMAP	(1 << PKMAP_ORDER)
-#ifndef CONFIG_PPC_4K_PAGES
-#define PKMAP_BASE	(FIXADDR_START - PAGE_SIZE*(LAST_PKMAP + 1))
-#else
-#define PKMAP_BASE	((FIXADDR_START - PAGE_SIZE*(LAST_PKMAP + 1)) & PMD_MASK)
-#endif
 #define LAST_PKMAP_MASK	(LAST_PKMAP-1)
 #define PKMAP_NR(virt)  ((virt-PKMAP_BASE) >> PAGE_SHIFT)
 #define PKMAP_ADDR(nr)  (PKMAP_BASE + ((nr) << PAGE_SHIFT))
arch/powerpc/include/asm/nohash/32/pgtable.h +21 -12
index 987a658b18e1..db4e530dd5f3 100644
@@ -69,6 +69,17 @@ extern int icache_44x_need_flush;
  * virtual space that goes below PKMAP and FIXMAP
  */
 #ifdef CONFIG_HIGHMEM
+#ifdef CONFIG_PPC_4K_PAGES
+#define PKMAP_ORDER	PTE_SHIFT
+#else
+#define PKMAP_ORDER	9
+#endif
+#define LAST_PKMAP	(1 << PKMAP_ORDER)
+#ifndef CONFIG_PPC_4K_PAGES
+#define PKMAP_BASE	(FIXADDR_START - PAGE_SIZE*(LAST_PKMAP + 1))
+#else
+#define PKMAP_BASE	((FIXADDR_START - PAGE_SIZE*(LAST_PKMAP + 1)) & PMD_MASK)
+#endif
 #define KVIRT_TOP	PKMAP_BASE
 #else
 #define KVIRT_TOP	(0xfe000000UL)	/* for now, could be FIXMAP_BASE ? */
@@ -80,10 +91,11 @@ extern int icache_44x_need_flush;
  * and ioremap space
  */
 #ifdef CONFIG_NOT_COHERENT_CACHE
-#define IOREMAP_TOP	((KVIRT_TOP - CONFIG_CONSISTENT_SIZE) & PAGE_MASK)
+#define IOREMAP_END	((KVIRT_TOP - CONFIG_CONSISTENT_SIZE) & PAGE_MASK)
 #else
-#define IOREMAP_TOP	KVIRT_TOP
+#define IOREMAP_END	KVIRT_TOP
 #endif
+#define IOREMAP_BASE	VMALLOC_BASE
 
 /*
  * Just any arbitrary offset to the start of the vmalloc VM area: the
@@ -94,21 +106,16 @@ extern int icache_44x_need_flush;
  * area for the same reason. ;)
  *
  * We no longer map larger than phys RAM with the BATs so we don't have
- * to worry about the VMALLOC_OFFSET causing problems.  We do have to worry
- * about clashes between our early calls to ioremap() that start growing down
- * from IOREMAP_TOP being run into the VM area allocations (growing upwards
- * from VMALLOC_START).  For this reason we have ioremap_bot to check when
- * we actually run into our mappings setup in the early boot with the VM
- * system.  This really does become a problem for machines with good amounts
- * of RAM.  -- Cort
+ * to worry about the VMALLOC_OFFSET causing problems.
  */
 #define VMALLOC_OFFSET (0x1000000) /* 16M */
 #ifdef PPC_PIN_SIZE
-#define VMALLOC_START (((_ALIGN((long)high_memory, PPC_PIN_SIZE) + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1)))
+#define VMALLOC_BASE (((_ALIGN((long)high_memory, PPC_PIN_SIZE) + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1)))
 #else
-#define VMALLOC_START ((((long)high_memory + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1)))
+#define VMALLOC_BASE ((((long)high_memory + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1)))
 #endif
-#define VMALLOC_END	ioremap_bot
+#define VMALLOC_START	ioremap_bot
+#define VMALLOC_END	IOREMAP_END
 
 /*
  * Bits in a linux-style PTE.  These match the bits in the
@@ -325,6 +332,8 @@ static inline int pte_young(pte_t pte)
 
 int map_kernel_page(unsigned long va, phys_addr_t pa, int flags);
 
+#include <asm/fixmap.h>
+
 #endif /* !__ASSEMBLY__ */
 
 #endif /* __ASM_POWERPC_NOHASH_32_PGTABLE_H */
arch/powerpc/mm/dma-noncoherent.c +1 -1
index 382528475433..d0a8fe74f5a0 100644
@@ -43,7 +43,7 @@
  * can be further configured for specific applications under
  * the "Advanced Setup" menu. -Matt
  */
-#define CONSISTENT_BASE		(IOREMAP_TOP)
+#define CONSISTENT_BASE		(IOREMAP_END)
 #define CONSISTENT_END 		(CONSISTENT_BASE + CONFIG_CONSISTENT_SIZE)
 #define CONSISTENT_OFFSET(x)	(((unsigned long)(x) - CONSISTENT_BASE) >> PAGE_SHIFT)
 
arch/powerpc/mm/dump_linuxpagetables.c +3 -3
index 876e2a3c79f2..6022adb899b7 100644
@@ -452,11 +452,11 @@ static void populate_markers(void)
 	address_markers[i++].start_address =  VMEMMAP_BASE;
 #endif
 #else /* !CONFIG_PPC64 */
+	address_markers[i++].start_address = IOREMAP_BASE;
 	address_markers[i++].start_address = ioremap_bot;
-	address_markers[i++].start_address = IOREMAP_TOP;
 #ifdef CONFIG_NOT_COHERENT_CACHE
-	address_markers[i++].start_address = IOREMAP_TOP;
-	address_markers[i++].start_address = IOREMAP_TOP +
+	address_markers[i++].start_address = IOREMAP_END;
+	address_markers[i++].start_address = IOREMAP_END +
 					     CONFIG_CONSISTENT_SIZE;
 #endif
 #ifdef CONFIG_HIGHMEM
arch/powerpc/mm/init_32.c +5 -1
index 3e59e5d64b01..7fb9e5a9852a 100644
@@ -172,7 +172,11 @@ void __init MMU_init(void)
 	mapin_ram();
 
 	/* Initialize early top-down ioremap allocator */
-	ioremap_bot = IOREMAP_TOP;
+	if (IS_ENABLED(CONFIG_HIGHMEM))
+		high_memory = (void *) __va(lowmem_end_addr);
+	else
+		high_memory = (void *) __va(memblock_end_of_DRAM());
+	ioremap_bot = IOREMAP_BASE;
 
 	if (ppc_md.progress)
 		ppc_md.progress("MMU:exit", 0x211);
arch/powerpc/mm/ioremap.c +10 -11
index 0c8f6113e0f3..a498e3cade9e 100644
@@ -28,11 +28,15 @@
 
 #include "mmu_decl.h"
 
-#ifdef CONFIG_PPC32
-
+#if defined(CONFIG_PPC_BOOK3S_64) || defined(CONFIG_PPC32)
 unsigned long ioremap_bot;
+#else
+unsigned long ioremap_bot = IOREMAP_BASE;
+#endif
 EXPORT_SYMBOL(ioremap_bot);	/* aka VMALLOC_END */
 
+#ifdef CONFIG_PPC32
+
 void __iomem *
 ioremap(phys_addr_t addr, unsigned long size)
 {
@@ -90,7 +94,7 @@ __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
 	/*
 	 * Choose an address to map it to.
 	 * Once the vmalloc system is running, we use it.
-	 * Before then, we use space going down from IOREMAP_TOP
+	 * Before then, we use space going up from IOREMAP_BASE
 	 * (ioremap_bot records where we're up to).
 	 */
 	p = addr & PAGE_MASK;
@@ -135,7 +139,8 @@ __ioremap_caller(phys_addr_t addr, unsigned long size, unsigned long flags,
 		area->phys_addr = p;
 		v = (unsigned long) area->addr;
 	} else {
-		v = (ioremap_bot -= size);
+		v = ioremap_bot;
+		ioremap_bot += size;
 	}
 
 	/*
@@ -164,19 +169,13 @@ void iounmap(volatile void __iomem *addr)
 	if (v_block_mapped((unsigned long)addr))
 		return;
 
-	if (addr > high_memory && (unsigned long) addr < ioremap_bot)
+	if ((unsigned long) addr >= ioremap_bot)
 		vunmap((void *) (PAGE_MASK & (unsigned long)addr));
 }
 EXPORT_SYMBOL(iounmap);
 
 #else
 
-#ifdef CONFIG_PPC_BOOK3S_64
-unsigned long ioremap_bot;
-#else /* !CONFIG_PPC_BOOK3S_64 */
-unsigned long ioremap_bot = IOREMAP_BASE;
-#endif
-
 /**
  * __ioremap_at - Low level function to establish the page tables
  *                for an IO mapping
arch/powerpc/mm/mem.c +4 -3
index c3c39b02b2ba..b680aa78a4ac 100644
@@ -345,8 +345,9 @@ void __init mem_init(void)
 #ifdef CONFIG_SWIOTLB
 	swiotlb_init(0);
 #endif
-
+#ifdef CONFIG_PPC64
 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
+#endif
 	set_max_mapnr(max_pfn);
 	free_all_bootmem();
 
@@ -383,10 +384,10 @@ void __init mem_init(void)
 #endif /* CONFIG_HIGHMEM */
 #ifdef CONFIG_NOT_COHERENT_CACHE
 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
-		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
+		IOREMAP_END, IOREMAP_END + CONFIG_CONSISTENT_SIZE);
 #endif /* CONFIG_NOT_COHERENT_CACHE */
 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
-		ioremap_bot, IOREMAP_TOP);
+		IOREMAP_BASE, ioremap_bot);
 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
 		VMALLOC_START, VMALLOC_END);
 #endif /* CONFIG_PPC32 */
Signature · 1 line
2.13.3
#4 lore raw

PATCH v3 4/14 powerpc: common ioremap functions.

In reply to Christophe Leroy

__ioremap(), ioremap(), ioremap_wc() et ioremap_prot() are very similar between PPC32 and PPC64, they can easily be made common.

_PAGE_WRITE equals to _PAGE_RW on PPC32
_PAGE_RO and _PAGE_HWWRITE are 0 on PPC64

iounmap() can also be made common by renaming the PPC32 iounmap() as __iounmap() then __iounmap() can also be made common to PPC32 and PPC64.

This patch also makes __ioremap_caller() common to PPC32 and PPC64

---
 arch/powerpc/include/asm/book3s/64/pgtable.h |   2 +
 arch/powerpc/include/asm/machdep.h           |   2 +-
 arch/powerpc/mm/ioremap.c                    | 190 ++++++---------------------
 3 files changed, 46 insertions(+), 148 deletions(-)
arch/powerpc/include/asm/book3s/64/pgtable.h +2 -0
index c233915abb68..2111dc9758cc 100644
@@ -17,6 +17,8 @@
 #define _PAGE_NA		0
 #define _PAGE_RO		0
 #define _PAGE_USER		0
+#define _PAGE_HWWRITE		0
+#define _PAGE_COHERENT		0
 
 #define _PAGE_EXEC		0x00001 /* execute permission */
 #define _PAGE_WRITE		0x00002 /* write access allowed */
arch/powerpc/include/asm/machdep.h +1 -1
index ffe7c71e1132..84d99ed82d5d 100644
@@ -33,11 +33,11 @@ struct pci_host_bridge;
 
 struct machdep_calls {
 	char		*name;
-#ifdef CONFIG_PPC64
 	void __iomem *	(*ioremap)(phys_addr_t addr, unsigned long size,
 				   unsigned long flags, void *caller);
 	void		(*iounmap)(volatile void __iomem *token);
 
+#ifdef CONFIG_PPC64
 #ifdef CONFIG_PM
 	void		(*iommu_save)(void);
 	void		(*iommu_restore)(void);
arch/powerpc/mm/ioremap.c +43 -147

230 lines not on this page · raw message

Signature · 1 line
2.13.3
#5 lore raw

PATCH v3 5/14 powerpc: use _ALIGN_DOWN macro for VMALLOC_BASE

In reply to Christophe Leroy

Use _ALIGN_DOWN macro instead of open coding in define of VMALLOC_BASE

---
 arch/powerpc/include/asm/nohash/32/pgtable.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/powerpc/include/asm/nohash/32/pgtable.h +2 -2
index db4e530dd5f3..810945afe677 100644
@@ -110,9 +110,9 @@ extern int icache_44x_need_flush;
  */
 #define VMALLOC_OFFSET (0x1000000) /* 16M */
 #ifdef PPC_PIN_SIZE
-#define VMALLOC_BASE (((_ALIGN((long)high_memory, PPC_PIN_SIZE) + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1)))
+#define VMALLOC_BASE _ALIGN_DOWN(_ALIGN((long)high_memory, PPC_PIN_SIZE) + VMALLOC_OFFSET, VMALLOC_OFFSET)
 #else
-#define VMALLOC_BASE ((((long)high_memory + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1)))
+#define VMALLOC_BASE _ALIGN_DOWN((long)high_memory + VMALLOC_OFFSET, VMALLOC_OFFSET)
 #endif
 #define VMALLOC_START	ioremap_bot
 #define VMALLOC_END	IOREMAP_END
Signature · 1 line
2.13.3
#6 lore raw

PATCH v3 6/14 powerpc/nohash32: allow setting GUARDED attribute in the PMD directly

In reply to Christophe Leroy

On the 8xx, the GUARDED attribute of the pages is managed in the L1 entry, therefore to avoid having to copy it into L1 entry at each TLB miss, we have to set it in the PMD

In order to allow this, this patch splits the VM alloc space in two parts, one for VM alloc and non Guarded IO, and one for Guarded IO.

---
 arch/powerpc/include/asm/nohash/32/pgalloc.h | 12 ++++++++++++
 arch/powerpc/include/asm/nohash/32/pgtable.h | 18 ++++++++++++++++--
 arch/powerpc/mm/dump_linuxpagetables.c       | 26 ++++++++++++++++++++++++--
 arch/powerpc/mm/ioremap.c                    | 13 ++++++++++---
 arch/powerpc/mm/mem.c                        |  9 +++++++++
 arch/powerpc/mm/pgtable_32.c                 | 28 +++++++++++++++++++++++++++-
 arch/powerpc/platforms/Kconfig.cputype       |  2 ++
 7 files changed, 100 insertions(+), 8 deletions(-)
arch/powerpc/include/asm/nohash/32/pgalloc.h +12 -0
index 29d37bd1f3b3..81c19d6460bd 100644
@@ -58,6 +58,14 @@ static inline void pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmdp,
 	*pmdp = __pmd(__pa(pte) | _PMD_PRESENT);
 }
 
+#ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
+static inline void pmd_populate_kernel_g(struct mm_struct *mm, pmd_t *pmdp,
+					 pte_t *pte)
+{
+	*pmdp = __pmd(__pa(pte) | _PMD_PRESENT | _PMD_GUARDED);
+}
+#endif
+
 static inline void pmd_populate(struct mm_struct *mm, pmd_t *pmdp,
 				pgtable_t pte_page)
 {
@@ -83,6 +91,10 @@ static inline void pmd_populate(struct mm_struct *mm, pmd_t *pmdp,
 #define pmd_pgtable(pmd) pmd_page(pmd)
 #endif
 
+#define pte_alloc_kernel_g(pmd, address)			\
+	((unlikely(pmd_none(*(pmd))) && __pte_alloc_kernel_g(pmd, address))? \
+		NULL: pte_offset_kernel(pmd, address))
+
 extern pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long addr);
 extern pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long addr);
 
arch/powerpc/include/asm/nohash/32/pgtable.h +16 -2
index 810945afe677..7873722198e1 100644
@@ -80,9 +80,14 @@ extern int icache_44x_need_flush;
 #else
 #define PKMAP_BASE	((FIXADDR_START - PAGE_SIZE*(LAST_PKMAP + 1)) & PMD_MASK)
 #endif
-#define KVIRT_TOP	PKMAP_BASE
+#define _KVIRT_TOP	PKMAP_BASE
 #else
-#define KVIRT_TOP	(0xfe000000UL)	/* for now, could be FIXMAP_BASE ? */
+#define _KVIRT_TOP	(0xfe000000UL)	/* for now, could be FIXMAP_BASE ? */
+#endif
+#ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
+#define KVIRT_TOP	_ALIGN_DOWN(_KVIRT_TOP, PGDIR_SIZE)
+#else
+#define KVIRT_TOP	_KVIRT_TOP
 #endif
 
 /*
@@ -95,7 +100,11 @@ extern int icache_44x_need_flush;
 #else
 #define IOREMAP_END	KVIRT_TOP
 #endif
+#ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
+#define IOREMAP_BASE	_ALIGN_UP(VMALLOC_BASE + (IOREMAP_END - VMALLOC_BASE) / 2, PGDIR_SIZE)
+#else
 #define IOREMAP_BASE	VMALLOC_BASE
+#endif
 
 /*
  * Just any arbitrary offset to the start of the vmalloc VM area: the
@@ -114,8 +123,13 @@ extern int icache_44x_need_flush;
 #else
 #define VMALLOC_BASE _ALIGN_DOWN((long)high_memory + VMALLOC_OFFSET, VMALLOC_OFFSET)
 #endif
+#ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
+#define VMALLOC_START	VMALLOC_BASE
+#define VMALLOC_END	IOREMAP_BASE
+#else
 #define VMALLOC_START	ioremap_bot
 #define VMALLOC_END	IOREMAP_END
+#endif
 
 /*
  * Bits in a linux-style PTE.  These match the bits in the
arch/powerpc/mm/dump_linuxpagetables.c +24 -2
index 6022adb899b7..cd3797be5e05 100644
@@ -74,9 +74,9 @@ struct addr_marker {
 
 static struct addr_marker address_markers[] = {
 	{ 0,	"Start of kernel VM" },
+#ifdef CONFIG_PPC64
 	{ 0,	"vmalloc() Area" },
 	{ 0,	"vmalloc() End" },
-#ifdef CONFIG_PPC64
 	{ 0,	"isa I/O start" },
 	{ 0,	"isa I/O end" },
 	{ 0,	"phb I/O start" },
@@ -85,8 +85,19 @@ static struct addr_marker address_markers[] = {
 	{ 0,	"I/O remap end" },
 	{ 0,	"vmemmap start" },
 #else
+#ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
+	{ 0,	"vmalloc() Area" },
+	{ 0,	"vmalloc() End" },
 	{ 0,	"Early I/O remap start" },
 	{ 0,	"Early I/O remap end" },
+	{ 0,	"I/O remap start" },
+	{ 0,	"I/O remap end" },
+#else
+	{ 0,	"Early I/O remap start" },
+	{ 0,	"Early I/O remap end" },
+	{ 0,	"vmalloc() I/O remap start" },
+	{ 0,	"vmalloc() I/O remap end" },
+#endif
 #ifdef CONFIG_NOT_COHERENT_CACHE
 	{ 0,	"Consistent mem start" },
 	{ 0,	"Consistent mem end" },
@@ -437,9 +448,9 @@ static void populate_markers(void)
 	int i = 0;
 
 	address_markers[i++].start_address = PAGE_OFFSET;
+#ifdef CONFIG_PPC64
 	address_markers[i++].start_address = VMALLOC_START;
 	address_markers[i++].start_address = VMALLOC_END;
-#ifdef CONFIG_PPC64
 	address_markers[i++].start_address = ISA_IO_BASE;
 	address_markers[i++].start_address = ISA_IO_END;
 	address_markers[i++].start_address = PHB_IO_BASE;
@@ -452,8 +463,19 @@ static void populate_markers(void)
 	address_markers[i++].start_address =  VMEMMAP_BASE;
 #endif
 #else /* !CONFIG_PPC64 */
+#ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
+	address_markers[i++].start_address = VMALLOC_START;
+	address_markers[i++].start_address = VMALLOC_END;
 	address_markers[i++].start_address = IOREMAP_BASE;
 	address_markers[i++].start_address = ioremap_bot;
+	address_markers[i++].start_address = ioremap_bot;
+	address_markers[i++].start_address = IOREMAP_END;
+#else
+	address_markers[i++].start_address = IOREMAP_BASE;
+	address_markers[i++].start_address = ioremap_bot;
+	address_markers[i++].start_address = ioremap_bot;
+	address_markers[i++].start_address = IOREMAP_END;
+#endif
 #ifdef CONFIG_NOT_COHERENT_CACHE
 	address_markers[i++].start_address = IOREMAP_END;
 	address_markers[i++].start_address = IOREMAP_END +
arch/powerpc/mm/ioremap.c +10 -3
index a140feec2f8a..3eecf82ff93e 100644
@@ -134,9 +134,16 @@ void __iomem * __ioremap_caller(phys_addr_t addr, unsigned long size,
 	if (slab_is_available()) {
 		struct vm_struct *area;
 
-		area = __get_vm_area_caller(size, VM_IOREMAP,
-					    ioremap_bot, IOREMAP_END,
-					    caller);
+#ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
+		if (!(flags & _PAGE_GUARDED))
+			area = __get_vm_area_caller(size, VM_IOREMAP,
+						    VMALLOC_START, VMALLOC_END,
+						    caller);
+		else
+#endif
+			area = __get_vm_area_caller(size, VM_IOREMAP,
+						    ioremap_bot, IOREMAP_END,
+						    caller);
 		if (area == NULL)
 			return NULL;
 
arch/powerpc/mm/mem.c +9 -0
index b680aa78a4ac..fd7af7af5b58 100644
@@ -386,10 +386,19 @@ void __init mem_init(void)
 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
 		IOREMAP_END, IOREMAP_END + CONFIG_CONSISTENT_SIZE);
 #endif /* CONFIG_NOT_COHERENT_CACHE */
+#ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
+	pr_info("  * 0x%08lx..0x%08lx  : ioremap\n",
+		ioremap_bot, IOREMAP_END);
 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
 		IOREMAP_BASE, ioremap_bot);
+	pr_info("  * 0x%08lx..0x%08lx  : vmalloc\n",
+		VMALLOC_START, VMALLOC_END);
+#else
 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
 		VMALLOC_START, VMALLOC_END);
+	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
+		IOREMAP_BASE, ioremap_bot);
+#endif
 #endif /* CONFIG_PPC32 */
 }
 
arch/powerpc/mm/pgtable_32.c +27 -1
index 54a5bc0767a9..3aa0c78db95d 100644
@@ -70,6 +70,27 @@ pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
 	return ptepage;
 }
 
+#ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
+int __pte_alloc_kernel_g(pmd_t *pmd, unsigned long address)
+{
+	pte_t *new = pte_alloc_one_kernel(&init_mm, address);
+	if (!new)
+		return -ENOMEM;
+
+	smp_wmb(); /* See comment in __pte_alloc */
+
+	spin_lock(&init_mm.page_table_lock);
+	if (likely(pmd_none(*pmd))) {	/* Has another populated it ? */
+		pmd_populate_kernel_g(&init_mm, pmd, new);
+		new = NULL;
+	}
+	spin_unlock(&init_mm.page_table_lock);
+	if (new)
+		pte_free_kernel(&init_mm, new);
+	return 0;
+}
+#endif
+
 int map_kernel_page(unsigned long va, phys_addr_t pa, int flags)
 {
 	pmd_t *pd;
@@ -79,7 +100,12 @@ int map_kernel_page(unsigned long va, phys_addr_t pa, int flags)
 	/* Use upper 10 bits of VA to index the first level map */
 	pd = pmd_offset(pud_offset(pgd_offset_k(va), va), va);
 	/* Use middle 10 bits of VA to index the second-level map */
-	pg = pte_alloc_kernel(pd, va);
+#ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
+	if (flags & _PAGE_GUARDED)
+		pg = pte_alloc_kernel_g(pd, va);
+	else
+#endif
+		pg = pte_alloc_kernel(pd, va);
 	if (pg != 0) {
 		err = 0;
 		/* The PTE should never be already set nor present in the
arch/powerpc/platforms/Kconfig.cputype +2 -0
index cc892dcfa114..ac7cab9c28e8 100644
@@ -323,6 +323,8 @@ config ARCH_ENABLE_HUGEPAGE_MIGRATION
 	def_bool y
 	depends on PPC_BOOK3S_64 && HUGETLB_PAGE && MIGRATION
 
+config PPC_GUARDED_PAGE_IN_PMD
+	bool
 
 config PPC_MMU_NOHASH
 	def_bool y
Signature · 1 line
2.13.3
#7 lore raw

PATCH v3 7/14 powerpc/8xx: set GUARDED attribute in the PMD directly

In reply to Christophe Leroy

On the 8xx, the GUARDED attribute of the pages is managed in the L1 entry, therefore to avoid having to copy it into L1 entry at each TLB miss, we set it in the PMD.

---
 arch/powerpc/include/asm/nohash/32/pte-8xx.h |  3 ++-
 arch/powerpc/kernel/head_8xx.S               | 18 +++++++-----------
 arch/powerpc/platforms/Kconfig.cputype       |  1 +
 3 files changed, 10 insertions(+), 12 deletions(-)
arch/powerpc/include/asm/nohash/32/pte-8xx.h +2 -1
index f04cb46ae8a1..a9a2919251e0 100644
@@ -47,10 +47,11 @@
 #define _PAGE_RO	0x0600	/* Supervisor RO, User no access */
 
 #define _PMD_PRESENT	0x0001
-#define _PMD_BAD	0x0fd0
+#define _PMD_BAD	0x0fc0
 #define _PMD_PAGE_MASK	0x000c
 #define _PMD_PAGE_8M	0x000c
 #define _PMD_PAGE_512K	0x0004
+#define _PMD_GUARDED	0x0010
 #define _PMD_USER	0x0020	/* APG 1 */
 
 /* Until my rework is finished, 8xx still needs atomic PTE updates */
arch/powerpc/kernel/head_8xx.S +7 -11
index 19bdc65d05b8..c310cc9ef489 100644
@@ -345,6 +345,10 @@ _ENTRY(ITLBMiss_cmp)
 	rlwinm	r10, r10, 32 - (PAGE_SHIFT - 2), 32 - PAGE_SHIFT, 29
 #ifdef CONFIG_HUGETLB_PAGE
 	mtcr	r11
+#endif
+	/* Load the MI_TWC with the attributes for this "segment." */
+	mtspr	SPRN_MI_TWC, r11	/* Set segment attributes */
+#ifdef CONFIG_HUGETLB_PAGE
 	bt-	28, 10f		/* bit 28 = Large page (8M) */
 	bt-	29, 20f		/* bit 29 = Large page (8M or 512k) */
 #endif
@@ -354,8 +358,6 @@ _ENTRY(ITLBMiss_cmp)
 #if defined(ITLB_MISS_KERNEL) || defined(CONFIG_HUGETLB_PAGE)
 	mtcr	r12
 #endif
-	/* Load the MI_TWC with the attributes for this "segment." */
-	mtspr	SPRN_MI_TWC, r11	/* Set segment attributes */
 
 #ifdef CONFIG_SWAP
 	rlwinm	r11, r10, 32-5, _PAGE_PRESENT
@@ -457,6 +459,9 @@ _ENTRY(DTLBMiss_jmp)
 	rlwinm	r10, r10, 32 - (PAGE_SHIFT - 2), 32 - PAGE_SHIFT, 29
 #ifdef CONFIG_HUGETLB_PAGE
 	mtcr	r11
+#endif
+	mtspr	SPRN_MD_TWC, r11
+#ifdef CONFIG_HUGETLB_PAGE
 	bt-	28, 10f		/* bit 28 = Large page (8M) */
 	bt-	29, 20f		/* bit 29 = Large page (8M or 512k) */
 #endif
@@ -465,15 +470,6 @@ _ENTRY(DTLBMiss_jmp)
 4:
 	mtcr	r12
 
-	/* Insert the Guarded flag into the TWC from the Linux PTE.
-	 * It is bit 27 of both the Linux PTE and the TWC (at least
-	 * I got that right :-).  It will be better when we can put
-	 * this into the Linux pgd/pmd and load it in the operation
-	 * above.
-	 */
-	rlwimi	r11, r10, 0, _PAGE_GUARDED
-	mtspr	SPRN_MD_TWC, r11
-
 	/* Both _PAGE_ACCESSED and _PAGE_PRESENT has to be set.
 	 * We also need to know if the insn is a load/store, so:
 	 * Clear _PAGE_PRESENT and load that which will
arch/powerpc/platforms/Kconfig.cputype +1 -0
index ac7cab9c28e8..09ed04377a98 100644
@@ -34,6 +34,7 @@ config PPC_8xx
 	bool "Freescale 8xx"
 	select FSL_SOC
 	select SYS_SUPPORTS_HUGETLBFS
+	select PPC_GUARDED_PAGE_IN_PMD
 
 config 40x
 	bool "AMCC 40x"
Signature · 1 line
2.13.3
#8 lore raw

PATCH v3 8/14 powerpc/8xx: Remove PTE_ATOMIC_UPDATES

In reply to Christophe Leroy

commit 1bc54c03117b9 ("powerpc: rework 4xx PTE access and TLB miss") introduced non atomic PTE updates and started the work of removing PTE updates in TLB miss handlers, but kept PTE_ATOMIC_UPDATES for the 8xx with the following comment:
/* Until my rework is finished, 8xx still needs atomic PTE updates */

commit fe11dc3f9628e ("powerpc/8xx: Update TLB asm so it behaves as linux mm expects") removed all PTE updates done in TLB miss handlers

Therefore, atomic PTE updates are not needed anymore for the 8xx

---
 arch/powerpc/include/asm/nohash/32/pte-8xx.h | 3 ---
 1 file changed, 3 deletions(-)
arch/powerpc/include/asm/nohash/32/pte-8xx.h +0 -3
index a9a2919251e0..31401320c1a5 100644
@@ -54,9 +54,6 @@
 #define _PMD_GUARDED	0x0010
 #define _PMD_USER	0x0020	/* APG 1 */
 
-/* Until my rework is finished, 8xx still needs atomic PTE updates */
-#define PTE_ATOMIC_UPDATES	1
-
 #ifdef CONFIG_PPC_16K_PAGES
 #define _PAGE_PSIZE	_PAGE_HUGE
 #endif
Signature · 1 line
2.13.3
#9 lore raw

PATCH v3 9/14 powerpc/mm: Use hardware assistance in TLB handlers on the 8xx

In reply to Christophe Leroy

Today, on the 8xx the TLB handlers do SW tablewalk by doing all the calculation in ASM, in order to match with the Linux page table structure.

The 8xx offers hardware assistance which allows significant size reduction of the TLB handlers, hence also reduces the time spent in the handlers.

However, using this HW assistance implies some constraints on the page table structure:
- Regardless of the main page size used (4k or 16k), the level 1 table (PGD) contains 1024 entries and each PGD entry covers a 4Mbytes area which is managed by a level 2 table (PTE) containing also 1024 entries each describing a 4k page.
- 16k pages require 4 identifical entries in the L2 table
- 512k pages PTE have to be spread every 128 bytes in the L2 table
- 8M pages PTE are at the address pointed by the L1 entry and each 8M page require 2 identical entries in the PGD.

In order to use hardware assistance, this patch does the following modifications:
- Make PGD size independant of the main page size
- In 16k pages mode, redefine pte_t as a struct with 4 elements, and populate those 4 elements in __set_pte_at() and pte_update()
- Modify the TLB handlers to use HW assistance
- Adapt the size of the hugepage tables.

Before that patch, the mean time spent in TLB miss handlers is:
- ITLB miss: 80 ticks
- DTLB miss: 62 ticks
After that patch, the mean time spent in TLB miss handlers is:
- ITLB miss: 72 ticks
- DTLB miss: 54 ticks
So the improvement is 10% for ITLB and 13% for DTLB misses

---
 arch/powerpc/include/asm/hugetlb.h           |   4 +-
 arch/powerpc/include/asm/nohash/32/pgtable.h |  16 +-
 arch/powerpc/include/asm/nohash/pgtable.h    |   4 +
 arch/powerpc/include/asm/pgtable-types.h     |   4 +
 arch/powerpc/kernel/head_8xx.S               | 225 +++++++++------------------
 arch/powerpc/mm/8xx_mmu.c                    |  10 +-
 arch/powerpc/mm/hugetlbpage.c                |  12 ++
 7 files changed, 111 insertions(+), 164 deletions(-)
arch/powerpc/include/asm/hugetlb.h +3 -1
index 96444bc08034..7c7d8351b566 100644
@@ -77,7 +77,9 @@ static inline pte_t *hugepte_offset(hugepd_t hpd, unsigned long addr,
 	unsigned long idx = 0;
 
 	pte_t *dir = hugepd_page(hpd);
-#ifndef CONFIG_PPC_FSL_BOOK3E
+#ifdef CONFIG_PPC_8xx
+	idx = (addr & ((1UL << pdshift) - 1)) >> PAGE_SHIFT;
+#elif !defined(CONFIG_PPC_FSL_BOOK3E)
 	idx = (addr & ((1UL << pdshift) - 1)) >> hugepd_shift(hpd);
 #endif
 
arch/powerpc/include/asm/nohash/32/pgtable.h +15 -1
index 7873722198e1..5872d79360a9 100644
@@ -18,7 +18,11 @@ extern int icache_44x_need_flush;
 
 #endif /* __ASSEMBLY__ */
 
+#if defined(CONFIG_PPC_8xx) && defined(CONFIG_PPC_16K_PAGES)
+#define PTE_INDEX_SIZE  (PTE_SHIFT - 2)
+#else
 #define PTE_INDEX_SIZE	PTE_SHIFT
+#endif
 #define PMD_INDEX_SIZE	0
 #define PUD_INDEX_SIZE	0
 #define PGD_INDEX_SIZE	(32 - PGDIR_SHIFT)
@@ -47,7 +51,11 @@ extern int icache_44x_need_flush;
  * -Matt
  */
 /* PGDIR_SHIFT determines what a top-level page table entry can map */
+#ifdef CONFIG_PPC_8xx
+#define PGDIR_SHIFT	22
+#else
 #define PGDIR_SHIFT	(PAGE_SHIFT + PTE_INDEX_SIZE)
+#endif
 #define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
 #define PGDIR_MASK	(~(PGDIR_SIZE-1))
 
@@ -201,7 +209,13 @@ static inline unsigned long pte_update(pte_t *p,
 	: "cc" );
 #else /* PTE_ATOMIC_UPDATES */
 	unsigned long old = pte_val(*p);
-	*p = __pte((old & ~clr) | set);
+	unsigned long new = (old & ~clr) | set;
+
+#if defined(CONFIG_PPC_8xx) && defined(CONFIG_PPC_16K_PAGES)
+	p->pte = p->pte1 = p->pte2 = p->pte3 = new;
+#else
+	*p = __pte(new);
+#endif
 #endif /* !PTE_ATOMIC_UPDATES */
 
 #ifdef CONFIG_44x
arch/powerpc/include/asm/nohash/pgtable.h +4 -0
index 2160be2e4339..bb5b65971b37 100644
@@ -164,7 +164,11 @@ static inline void __set_pte_at(struct mm_struct *mm, unsigned long addr,
 	/* Anything else just stores the PTE normally. That covers all 64-bit
 	 * cases, and 32-bit non-hash with 32-bit PTEs.
 	 */
+#if defined(CONFIG_PPC_8xx) && defined(CONFIG_PPC_16K_PAGES)
+	ptep->pte = ptep->pte1 = ptep->pte2 = ptep->pte3 = pte_val(pte);
+#else
 	*ptep = pte;
+#endif
 
 	/*
 	 * With hardware tablewalk, a sync is needed to ensure that
arch/powerpc/include/asm/pgtable-types.h +4 -0
index eccb30b38b47..3b0edf041b2e 100644
@@ -3,7 +3,11 @@
 #define _ASM_POWERPC_PGTABLE_TYPES_H
 
 /* PTE level */
+#if defined(CONFIG_PPC_8xx) && defined(CONFIG_PPC_16K_PAGES)
+typedef struct { pte_basic_t pte, pte1, pte2, pte3; } pte_t;
+#else
 typedef struct { pte_basic_t pte; } pte_t;
+#endif
 #define __pte(x)	((pte_t) { (x) })
 static inline pte_basic_t pte_val(pte_t x)
 {
arch/powerpc/kernel/head_8xx.S +71 -154

409 lines not on this page · raw message

arch/powerpc/mm/8xx_mmu.c +2 -8
index 5d53684c2ebd..54a02b8e21ec 100644
@@ -173,8 +173,6 @@ void __init setup_initial_memory_limit(phys_addr_t first_memblock_base,
  */
 void set_context(unsigned long id, pgd_t *pgd)
 {
-	s16 offset = (s16)(__pa(swapper_pg_dir));
-
 #ifdef CONFIG_BDI_SWITCH
 	pgd_t	**ptr = *(pgd_t ***)(KERNELBASE + 0xf0);
 
@@ -184,12 +182,8 @@ void set_context(unsigned long id, pgd_t *pgd)
 	*(ptr + 1) = pgd;
 #endif
 
-	/* Register M_TW will contain base address of level 1 table minus the
-	 * lower part of the kernel PGDIR base address, so that all accesses to
-	 * level 1 table are done relative to lower part of kernel PGDIR base
-	 * address.
-	 */
-	mtspr(SPRN_M_TW, __pa(pgd) - offset);
+	/* Register M_TWB will contain base address of level 1 table */
+	mtspr(SPRN_M_TWB, __pa(pgd));
 
 	/* Update context */
 	mtspr(SPRN_M_CASID, id - 1);
arch/powerpc/mm/hugetlbpage.c +12 -0
index 2a4b1bf8bde6..d889196bbaf6 100644
@@ -63,7 +63,11 @@ static int __hugepte_alloc(struct mm_struct *mm, hugepd_t *hpdp,
 		cachep = hugepte_cache;
 		num_hugepd = 1 << (pshift - pdshift);
 	} else {
+#ifdef CONFIG_PPC_8xx
+		cachep = PGT_CACHE(PTE_SHIFT);
+#else
 		cachep = PGT_CACHE(pdshift - pshift);
+#endif
 		num_hugepd = 1;
 	}
 
@@ -328,7 +332,11 @@ static void free_hugepd_range(struct mmu_gather *tlb, hugepd_t *hpdp, int pdshif
 	if (shift >= pdshift)
 		hugepd_free(tlb, hugepte);
 	else
+#ifdef CONFIG_PPC_8xx
+		pgtable_free_tlb(tlb, hugepte, PTE_SHIFT);
+#else
 		pgtable_free_tlb(tlb, hugepte, pdshift - shift);
+#endif
 }
 
 static void hugetlb_free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
@@ -696,7 +704,11 @@ static int __init hugetlbpage_init(void)
 		 * use pgt cache for hugepd.
 		 */
 		if (pdshift > shift)
+#ifdef CONFIG_PPC_8xx
+			pgtable_cache_add(PTE_SHIFT, NULL);
+#else
 			pgtable_cache_add(pdshift - shift, NULL);
+#endif
 #if defined(CONFIG_PPC_FSL_BOOK3E) || defined(CONFIG_PPC_8xx)
 		else if (!hugepte_cache) {
 			/*
Signature · 1 line
2.13.3
#10 lore raw

PATCH v3 10/14 powerpc/8xx: reunify TLB handler routines

In reply to Christophe Leroy

Each handler must not exceed 64 instructions to fit into the main exception area.
Following the significant size reduction of TLB handler routines, the side handlers can be brought back close to the main part.

In the worst case:
Main part of ITLB handler is 45 insn, side part is 9 insn ==> total 54 Main part of DTLB handler is 37 insn, side part is 23 insn ==> total 60

---
 arch/powerpc/kernel/head_8xx.S | 108 ++++++++++++++++++++---------------------
 1 file changed, 52 insertions(+), 56 deletions(-)
arch/powerpc/kernel/head_8xx.S +52 -56
index 15ad8bda676a..9035f3ba010f 100644
@@ -388,6 +388,23 @@ _ENTRY(itlb_miss_perf)
 	mfspr	r11, SPRN_SPRG_SCRATCH1
 	rfi
 
+#ifndef CONFIG_PIN_TLB_TEXT
+ITLBMissLinear:
+	mtcr	r11
+	/* Set 8M byte page and mark it valid */
+	li	r11, MI_PS8MEG | MI_SVALID
+	mtspr	SPRN_MI_TWC, r11
+	rlwinm	r10, r10, 20, 0x0f800000	/* 8xx supports max 256Mb RAM */
+	ori	r10, r10, 0xf0 | MI_SPS16K | _PAGE_PRIVILEGED | _PAGE_DIRTY | \
+			  _PAGE_PRESENT
+	mtspr	SPRN_MI_RPN, r10	/* Update TLB entry */
+
+_ENTRY(itlb_miss_exit_2)
+	mfspr	r10, SPRN_SPRG_SCRATCH0
+	mfspr	r11, SPRN_SPRG_SCRATCH1
+	rfi
+#endif
+
 	. = 0x1200
 DataStoreTLBMiss:
 	mtspr	SPRN_SPRG_SCRATCH0, r10
@@ -463,6 +480,41 @@ _ENTRY(dtlb_miss_perf)
 	mfspr	r11, SPRN_SPRG_SCRATCH1
 	rfi
 
+DTLBMissIMMR:
+	mtcr	r11
+	/* Set 512k byte guarded page and mark it valid */
+	li	r10, MD_PS512K | MD_GUARDED | MD_SVALID
+	mtspr	SPRN_MD_TWC, r10
+	mfspr	r10, SPRN_IMMR			/* Get current IMMR */
+	rlwinm	r10, r10, 0, 0xfff80000		/* Get 512 kbytes boundary */
+	ori	r10, r10, 0xf0 | MD_SPS16K | _PAGE_PRIVILEGED | _PAGE_DIRTY | \
+			  _PAGE_PRESENT | _PAGE_NO_CACHE
+	mtspr	SPRN_MD_RPN, r10	/* Update TLB entry */
+
+	li	r11, RPN_PATTERN
+	mtspr	SPRN_DAR, r11	/* Tag DAR */
+_ENTRY(dtlb_miss_exit_2)
+	mfspr	r10, SPRN_SPRG_SCRATCH0
+	mfspr	r11, SPRN_SPRG_SCRATCH1
+	rfi
+
+DTLBMissLinear:
+	mtcr	r11
+	/* Set 8M byte page and mark it valid */
+	li	r11, MD_PS8MEG | MD_SVALID
+	mtspr	SPRN_MD_TWC, r11
+	rlwinm	r10, r10, 20, 0x0f800000	/* 8xx supports max 256Mb RAM */
+	ori	r10, r10, 0xf0 | MD_SPS16K | _PAGE_PRIVILEGED | _PAGE_DIRTY | \
+			  _PAGE_PRESENT
+	mtspr	SPRN_MD_RPN, r10	/* Update TLB entry */
+
+	li	r11, RPN_PATTERN
+	mtspr	SPRN_DAR, r11	/* Tag DAR */
+_ENTRY(dtlb_miss_exit_3)
+	mfspr	r10, SPRN_SPRG_SCRATCH0
+	mfspr	r11, SPRN_SPRG_SCRATCH1
+	rfi
+
 /* This is an instruction TLB error on the MPC8xx.  This could be due
  * to many reasons, such as executing guarded memory or illegal instruction
  * addresses.  There is nothing to do but handle a big time error fault.
@@ -565,62 +617,6 @@ InstructionBreakpoint:
 
 	. = 0x2000
 
-/*
- * Bottom part of DataStoreTLBMiss handlers for IMMR area and linear RAM.
- * not enough space in the DataStoreTLBMiss area.
- */
-DTLBMissIMMR:
-	mtcr	r11
-	/* Set 512k byte guarded page and mark it valid */
-	li	r10, MD_PS512K | MD_GUARDED | MD_SVALID
-	mtspr	SPRN_MD_TWC, r10
-	mfspr	r10, SPRN_IMMR			/* Get current IMMR */
-	rlwinm	r10, r10, 0, 0xfff80000		/* Get 512 kbytes boundary */
-	ori	r10, r10, 0xf0 | MD_SPS16K | _PAGE_PRIVILEGED | _PAGE_DIRTY | \
-			  _PAGE_PRESENT | _PAGE_NO_CACHE
-	mtspr	SPRN_MD_RPN, r10	/* Update TLB entry */
-
-	li	r11, RPN_PATTERN
-	mtspr	SPRN_DAR, r11	/* Tag DAR */
-_ENTRY(dtlb_miss_exit_2)
-	mfspr	r10, SPRN_SPRG_SCRATCH0
-	mfspr	r11, SPRN_SPRG_SCRATCH1
-	rfi
-
-DTLBMissLinear:
-	mtcr	r11
-	/* Set 8M byte page and mark it valid */
-	li	r11, MD_PS8MEG | MD_SVALID
-	mtspr	SPRN_MD_TWC, r11
-	rlwinm	r10, r10, 20, 0x0f800000	/* 8xx supports max 256Mb RAM */
-	ori	r10, r10, 0xf0 | MD_SPS16K | _PAGE_PRIVILEGED | _PAGE_DIRTY | \
-			  _PAGE_PRESENT
-	mtspr	SPRN_MD_RPN, r10	/* Update TLB entry */
-
-	li	r11, RPN_PATTERN
-	mtspr	SPRN_DAR, r11	/* Tag DAR */
-_ENTRY(dtlb_miss_exit_3)
-	mfspr	r10, SPRN_SPRG_SCRATCH0
-	mfspr	r11, SPRN_SPRG_SCRATCH1
-	rfi
-
-#ifndef CONFIG_PIN_TLB_TEXT
-ITLBMissLinear:
-	mtcr	r11
-	/* Set 8M byte page and mark it valid */
-	li	r11, MI_PS8MEG | MI_SVALID
-	mtspr	SPRN_MI_TWC, r11
-	rlwinm	r10, r10, 20, 0x0f800000	/* 8xx supports max 256Mb RAM */
-	ori	r10, r10, 0xf0 | MI_SPS16K | _PAGE_PRIVILEGED | _PAGE_DIRTY | \
-			  _PAGE_PRESENT
-	mtspr	SPRN_MI_RPN, r10	/* Update TLB entry */
-
-_ENTRY(itlb_miss_exit_2)
-	mfspr	r10, SPRN_SPRG_SCRATCH0
-	mfspr	r11, SPRN_SPRG_SCRATCH1
-	rfi
-#endif
-
 /* This is the procedure to calculate the data EA for buggy dcbx,dcbi instructions
  * by decoding the registers used by the dcbx instruction and adding them.
  * DAR is set to the calculated address.
Signature · 1 line
2.13.3
#11 lore raw

PATCH v3 11/14 powerpc/8xx: Free up SPRN_SPRG_SCRATCH2

In reply to Christophe Leroy

We can now use SPRN_M_TW in the DAR Fixup code, freeing SPRN_SPRG_SCRATCH2

Then SPRN_SPRG_SCRATCH2 may be used for something else in the future.

---
 arch/powerpc/kernel/head_8xx.S | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
arch/powerpc/kernel/head_8xx.S +4 -4
index 9035f3ba010f..81b125475213 100644
@@ -624,7 +624,7 @@ InstructionBreakpoint:
  /* define if you don't want to use self modifying code */
 #define NO_SELF_MODIFYING_CODE
 FixupDAR:/* Entry point for dcbx workaround. */
-	mtspr	SPRN_SPRG_SCRATCH2, r10
+	mtspr	SPRN_M_TW, r10
 	/* fetch instruction from memory. */
 	mfspr	r10, SPRN_SRR0
 	mtspr	SPRN_MD_EPN, r10
@@ -668,7 +668,7 @@ _ENTRY(FixupDAR_cmp)
 	beq+	142f
 	cmpwi	cr0, r10, 1964	/* Is icbi? */
 	beq+	142f
-141:	mfspr	r10,SPRN_SPRG_SCRATCH2
+141:	mfspr	r10,SPRN_M_TW
 	b	DARFixed	/* Nope, go back to normal TLB processing */
 
 200:
@@ -703,7 +703,7 @@ modified_instr:
 	bne+	143f
 	subf	r10,r0,r10	/* r10=r10-r0, only if reg RA is r0 */
 143:	mtdar	r10		/* store faulting EA in DAR */
-	mfspr	r10,SPRN_SPRG_SCRATCH2
+	mfspr	r10,SPRN_M_TW
 	b	DARFixed	/* Go back to normal TLB handling */
 #else
 	mfctr	r10
@@ -757,7 +757,7 @@ modified_instr:
 	mfdar	r11
 	mtctr	r11			/* restore ctr reg from DAR */
 	mtdar	r10			/* save fault EA to DAR */
-	mfspr	r10,SPRN_SPRG_SCRATCH2
+	mfspr	r10,SPRN_M_TW
 	b	DARFixed		/* Go back to normal TLB handling */
 
 	/* special handling for r10,r11 since these are modified already */
Signature · 1 line
2.13.3
#12 lore raw

PATCH v3 12/14 powerpc/mm: Make pte_fragment_alloc() common to PPC32 and PPC64

In reply to Christophe Leroy

In order to allow the 8xx to handle pte_fragments, this patch makes it common to PPC32 and PPC64 by moving the related code to common files and by defining a new config item called CONFIG_NEED_PTE_FRAG

---
 arch/powerpc/include/asm/mmu_context.h | 53 ++++++++++++++++++++++++
 arch/powerpc/include/asm/page.h        |  2 +-
 arch/powerpc/mm/mmu_context_book3s64.c | 44 --------------------
 arch/powerpc/mm/pgtable-book3s64.c     | 72 ---------------------------------
 arch/powerpc/mm/pgtable.c              | 74 ++++++++++++++++++++++++++++++++++
 arch/powerpc/platforms/Kconfig.cputype | 15 +++++++
 6 files changed, 143 insertions(+), 117 deletions(-)
arch/powerpc/include/asm/mmu_context.h +53 -0
index 1835ca1505d6..aa098b640f74 100644
@@ -262,5 +262,58 @@ static inline u64 pte_to_hpte_pkey_bits(u64 pteflags)
 
 #endif /* CONFIG_PPC_MEM_KEYS */
 
+#ifdef CONFIG_NEED_PTE_FRAG
+static inline void pte_frag_destroy(void *pte_frag)
+{
+	int count;
+	struct page *page;
+
+	page = virt_to_page(pte_frag);
+	/* drop all the pending references */
+	count = ((unsigned long)pte_frag & ~PAGE_MASK) >> PTE_FRAG_SIZE_SHIFT;
+	/* We allow PTE_FRAG_NR fragments from a PTE page */
+	if (page_ref_sub_and_test(page, PTE_FRAG_NR - count)) {
+		pgtable_page_dtor(page);
+		free_unref_page(page);
+	}
+}
+#endif
+
+#ifdef CONFIG_NEED_PMD_FRAG
+static inline void pmd_frag_destroy(void *pmd_frag)
+{
+	int count;
+	struct page *page;
+
+	page = virt_to_page(pmd_frag);
+	/* drop all the pending references */
+	count = ((unsigned long)pmd_frag & ~PAGE_MASK) >> PMD_FRAG_SIZE_SHIFT;
+	/* We allow PTE_FRAG_NR fragments from a PTE page */
+	if (page_ref_sub_and_test(page, PMD_FRAG_NR - count)) {
+		pgtable_pmd_page_dtor(page);
+		free_unref_page(page);
+	}
+}
+#endif
+
+static inline void destroy_pagetable_page(struct mm_struct *mm)
+{
+#ifdef CONFIG_NEED_PAGE_FRAG
+	void *frag;
+
+#ifdef CONFIG_NEED_PTE_FRAG
+	frag = mm->context.pte_frag;
+	if (frag)
+		pte_frag_destroy(frag);
+#endif
+
+#ifdef CONFIG_NEED_PMD_FRAG
+	frag = mm->context.pmd_frag;
+	if (frag)
+		pmd_frag_destroy(frag);
+#endif
+#endif
+}
+
 #endif /* __KERNEL__ */
 #endif /* __ASM_POWERPC_MMU_CONTEXT_H */
arch/powerpc/include/asm/page.h +1 -1
index db7be0779d55..cd10564ff7df 100644
@@ -344,7 +344,7 @@ struct vm_area_struct;
  */
 typedef pte_t *pgtable_t;
 #else
-#if defined(CONFIG_PPC_64K_PAGES) && defined(CONFIG_PPC64)
+#if defined(CONFIG_NEED_PTE_FRAG)
 typedef pte_t *pgtable_t;
 #else
 typedef struct page *pgtable_t;
arch/powerpc/mm/mmu_context_book3s64.c +0 -44

50 lines not on this page · raw message

arch/powerpc/mm/pgtable-book3s64.c +0 -72

78 lines not on this page · raw message

arch/powerpc/mm/pgtable.c +74 -0

77 lines not on this page · raw message

arch/powerpc/platforms/Kconfig.cputype +15 -0
index 09ed04377a98..ddccae5ed192 100644
@@ -341,6 +341,21 @@ config PPC_MM_SLICES
 	default y if PPC_8xx && HUGETLB_PAGE
 	default n
 
+config NEED_PAGE_FRAG
+	bool
+
+config NEED_PTE_FRAG
+	bool
+	default y if PPC_BOOK3S_64
+	default n
+	select NEED_PAGE_FRAG
+
+config NEED_PMD_FRAG
+	bool
+	default y if PPC_BOOK3S_64
+	default n
+	select NEED_PAGE_FRAG
+
 config PPC_HAVE_PMU_SUPPORT
        bool
 
Signature · 1 line
2.13.3
#13 lore raw

PATCH v3 13/14 powerpc/mm: Use pte_fragment_alloc() on 8xx

In reply to Christophe Leroy

In 16k page size mode, the 8xx need only 4k for a page table.

This patch makes use of the pte_fragment functions in order to avoid wasting memory space

---
 arch/powerpc/include/asm/mmu-8xx.h           |  4 +++
 arch/powerpc/include/asm/nohash/32/pgalloc.h | 44 +++++++++++++++++++++++++++-
 arch/powerpc/include/asm/nohash/32/pgtable.h | 10 ++++++-
 arch/powerpc/mm/mmu_context_nohash.c         |  4 +++
 arch/powerpc/mm/pgtable.c                    | 10 ++++++-
 arch/powerpc/mm/pgtable_32.c                 | 12 ++++++++
 arch/powerpc/platforms/Kconfig.cputype       |  1 +
 7 files changed, 82 insertions(+), 3 deletions(-)
arch/powerpc/include/asm/mmu-8xx.h +4 -0
index 193f53116c7a..4f4cb754afd8 100644
@@ -190,6 +190,10 @@ typedef struct {
 	struct slice_mask mask_8m;
 # endif
 #endif
+#ifdef CONFIG_NEED_PTE_FRAG
+	/* for 4K PTE fragment support */
+	void *pte_frag;
+#endif
 } mm_context_t;
 
 #define PHYS_IMMR_BASE (mfspr(SPRN_IMMR) & 0xfff80000)
arch/powerpc/include/asm/nohash/32/pgalloc.h +43 -1

78 lines not on this page · raw message

arch/powerpc/include/asm/nohash/32/pgtable.h +9 -1
index 5872d79360a9..ab5671e14fa1 100644
@@ -20,6 +20,9 @@ extern int icache_44x_need_flush;
 
 #if defined(CONFIG_PPC_8xx) && defined(CONFIG_PPC_16K_PAGES)
 #define PTE_INDEX_SIZE  (PTE_SHIFT - 2)
+#define PTE_FRAG_NR		4
+#define PTE_FRAG_SIZE_SHIFT	12
+#define PTE_FRAG_SIZE		(1UL << PTE_FRAG_SIZE_SHIFT)
 #else
 #define PTE_INDEX_SIZE	PTE_SHIFT
 #endif
@@ -319,7 +322,7 @@ static inline int pte_young(pte_t pte)
  */
 #ifndef CONFIG_BOOKE
 #define pmd_page_vaddr(pmd)	\
-	((unsigned long) __va(pmd_val(pmd) & PAGE_MASK))
+	((unsigned long) __va(pmd_val(pmd) & ~(PTE_TABLE_SIZE - 1)))
 #define pmd_page(pmd)		\
 	pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT)
 #else
@@ -342,9 +345,14 @@ static inline int pte_young(pte_t pte)
 #define pte_offset_kernel(dir, addr)	\
 	(pmd_bad(*(dir)) ? NULL : (pte_t *)pmd_page_vaddr(*(dir)) + \
 				  pte_index(addr))
+#ifdef CONFIG_NEED_PTE_FRAG
+#define pte_offset_map(dir, addr)	pte_offset_kernel(dir, addr)
+#define pte_unmap(pte)		do {} while (0)
+#else
 #define pte_offset_map(dir, addr)		\
 	((pte_t *) kmap_atomic(pmd_page(*(dir))) + pte_index(addr))
 #define pte_unmap(pte)		kunmap_atomic(pte)
+#endif
 
 /*
  * Encode and decode a swap entry.
arch/powerpc/mm/mmu_context_nohash.c +4 -0
index be8f5c9d4d08..2b048e3addff 100644
@@ -344,6 +344,9 @@ int init_new_context(struct task_struct *t, struct mm_struct *mm)
 #endif
 	mm->context.id = MMU_NO_CONTEXT;
 	mm->context.active = 0;
+#ifdef CONFIG_NEED_PTE_FRAG
+	mm->context.pte_frag = NULL;
+#endif
 	return 0;
 }
 
@@ -372,6 +375,7 @@ void destroy_context(struct mm_struct *mm)
 		nr_free_contexts++;
 	}
 	raw_spin_unlock_irqrestore(&context_lock, flags);
+	destroy_pagetable_page(mm);
 }
 
 #ifdef CONFIG_SMP
arch/powerpc/mm/pgtable.c +9 -1
index 3b3e51e9e431..18a5cbf1d1a0 100644
@@ -23,6 +23,7 @@
 
 #include <linux/kernel.h>
 #include <linux/gfp.h>
+#include <linux/memblock.h>
 #include <linux/mm.h>
 #include <linux/percpu.h>
 #include <linux/hardirq.h>
@@ -327,10 +328,17 @@ static pte_t *__alloc_for_ptecache(struct mm_struct *mm, int kernel)
 	return (pte_t *)ret;
 }
 
-pte_t *pte_fragment_alloc(struct mm_struct *mm, unsigned long vmaddr, int kernel)
+__ref pte_t *pte_fragment_alloc(struct mm_struct *mm, unsigned long vmaddr, int kernel)
 {
 	pte_t *pte;
 
+	if (kernel && !slab_is_available()) {
+		pte = __va(memblock_alloc(PTE_FRAG_SIZE, PTE_FRAG_SIZE));
+		if (pte)
+			memset(pte, 0, PTE_FRAG_SIZE);
+
+		return pte;
+	}
 	pte = get_pte_from_cache(mm);
 	if (pte)
 		return pte;
arch/powerpc/mm/pgtable_32.c +12 -0
index 3aa0c78db95d..5c8737cf2945 100644
@@ -40,6 +40,17 @@
 
 extern char etext[], _stext[], _sinittext[], _einittext[];
 
+#ifdef CONFIG_NEED_PTE_FRAG
+void pte_fragment_free(unsigned long *table, int kernel)
+{
+	struct page *page = virt_to_page(table);
+	if (put_page_testzero(page)) {
+		if (!kernel)
+			pgtable_page_dtor(page);
+		free_unref_page(page);
+	}
+}
+#else
 __ref pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 {
 	pte_t *pte;
@@ -69,6 +80,7 @@ pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
 	}
 	return ptepage;
 }
+#endif
 
 #ifdef CONFIG_PPC_GUARDED_PAGE_IN_PMD
 int __pte_alloc_kernel_g(pmd_t *pmd, unsigned long address)
arch/powerpc/platforms/Kconfig.cputype +1 -0
index ddccae5ed192..482609732bc6 100644
@@ -347,6 +347,7 @@ config NEED_PAGE_FRAG
 config NEED_PTE_FRAG
 	bool
 	default y if PPC_BOOK3S_64
+	default y if PPC_8xx && PPC_16K_PAGES
 	default n
 	select NEED_PAGE_FRAG
 
Signature · 1 line
2.13.3
#14 lore raw

PATCH v3 14/14 powerpc/8xx: Move SW perf counters in first 32kb of memory

In reply to Christophe Leroy

In order to simplify time critical exceptions handling 8xx specific SW perf counters, this patch moves the counters into the begining of memory. This is possible because .text is readable and the counters are never modified outside of the handlers.

By doing this, we avoid having to set a second register with the upper part of the address of the counters.

---
 arch/powerpc/kernel/head_8xx.S | 75 +++++++++++++++++++-----------------------
 1 file changed, 34 insertions(+), 41 deletions(-)
arch/powerpc/kernel/head_8xx.S +34 -41
index 81b125475213..084a76a012b5 100644
@@ -106,6 +106,23 @@ turn_on_mmu:
 	mtspr	SPRN_SRR0,r0
 	rfi				/* enables MMU */
 
+
+#ifdef CONFIG_PERF_EVENTS
+	.align	4
+
+	.globl	itlb_miss_counter
+itlb_miss_counter:
+	.space	4
+
+	.globl	dtlb_miss_counter
+dtlb_miss_counter:
+	.space	4
+
+	.globl	instruction_counter
+instruction_counter:
+	.space	4
+#endif
+
 /*
  * Exception entry code.  This code runs with address translation
  * turned off, i.e. using physical addresses.
@@ -368,25 +385,20 @@ _ENTRY(ITLBMiss_cmp)
 	mtspr	SPRN_MI_RPN, r10	/* Update TLB entry */
 
 	/* Restore registers */
-_ENTRY(itlb_miss_exit_1)
-	mfspr	r10, SPRN_SPRG_SCRATCH0
 #if defined(ITLB_MISS_KERNEL) || defined(CONFIG_SWAP)
 	mfspr	r11, SPRN_SPRG_SCRATCH1
 #endif
+_ENTRY(itlb_miss_exit_1)
+	mfspr	r10, SPRN_SPRG_SCRATCH0
 	rfi
 #ifdef CONFIG_PERF_EVENTS
 _ENTRY(itlb_miss_perf)
-#if !defined(ITLB_MISS_KERNEL) && !defined(CONFIG_SWAP)
-	mtspr	SPRN_SPRG_SCRATCH1, r11
-#endif
-	lis	r10, (itlb_miss_counter - PAGE_OFFSET)@ha
-	lwz	r11, (itlb_miss_counter - PAGE_OFFSET)@l(r10)
-	addi	r11, r11, 1
-	stw	r11, (itlb_miss_counter - PAGE_OFFSET)@l(r10)
-#endif
+	lwz	r10, (itlb_miss_counter - PAGE_OFFSET)@l(0)
+	addi	r10, r10, 1
+	stw	r10, (itlb_miss_counter - PAGE_OFFSET)@l(0)
 	mfspr	r10, SPRN_SPRG_SCRATCH0
-	mfspr	r11, SPRN_SPRG_SCRATCH1
 	rfi
+#endif
 
 #ifndef CONFIG_PIN_TLB_TEXT
 ITLBMissLinear:
@@ -399,9 +411,9 @@ ITLBMissLinear:
 			  _PAGE_PRESENT
 	mtspr	SPRN_MI_RPN, r10	/* Update TLB entry */
 
+	mfspr	r11, SPRN_SPRG_SCRATCH1
 _ENTRY(itlb_miss_exit_2)
 	mfspr	r10, SPRN_SPRG_SCRATCH0
-	mfspr	r11, SPRN_SPRG_SCRATCH1
 	rfi
 #endif
 
@@ -465,20 +477,18 @@ _ENTRY(DTLBMiss_jmp)
 
 	/* Restore registers */
 	mtspr	SPRN_DAR, r11	/* Tag DAR */
+	mfspr	r11, SPRN_SPRG_SCRATCH1
 _ENTRY(dtlb_miss_exit_1)
 	mfspr	r10, SPRN_SPRG_SCRATCH0
-	mfspr	r11, SPRN_SPRG_SCRATCH1
 	rfi
 #ifdef CONFIG_PERF_EVENTS
 _ENTRY(dtlb_miss_perf)
-	lis	r10, (dtlb_miss_counter - PAGE_OFFSET)@ha
-	lwz	r11, (dtlb_miss_counter - PAGE_OFFSET)@l(r10)
-	addi	r11, r11, 1
-	stw	r11, (dtlb_miss_counter - PAGE_OFFSET)@l(r10)
-#endif
+	lwz	r10, (dtlb_miss_counter - PAGE_OFFSET)@l(0)
+	addi	r10, r10, 1
+	stw	r10, (dtlb_miss_counter - PAGE_OFFSET)@l(0)
 	mfspr	r10, SPRN_SPRG_SCRATCH0
-	mfspr	r11, SPRN_SPRG_SCRATCH1
 	rfi
+#endif
 
 DTLBMissIMMR:
 	mtcr	r11
@@ -493,9 +503,9 @@ DTLBMissIMMR:
 
 	li	r11, RPN_PATTERN
 	mtspr	SPRN_DAR, r11	/* Tag DAR */
+	mfspr	r11, SPRN_SPRG_SCRATCH1
 _ENTRY(dtlb_miss_exit_2)
 	mfspr	r10, SPRN_SPRG_SCRATCH0
-	mfspr	r11, SPRN_SPRG_SCRATCH1
 	rfi
 
 DTLBMissLinear:
@@ -510,9 +520,9 @@ DTLBMissLinear:
 
 	li	r11, RPN_PATTERN
 	mtspr	SPRN_DAR, r11	/* Tag DAR */
+	mfspr	r11, SPRN_SPRG_SCRATCH1
 _ENTRY(dtlb_miss_exit_3)
 	mfspr	r10, SPRN_SPRG_SCRATCH0
-	mfspr	r11, SPRN_SPRG_SCRATCH1
 	rfi
 
 /* This is an instruction TLB error on the MPC8xx.  This could be due
@@ -598,16 +608,13 @@ DataBreakpoint:
 	. = 0x1d00
 InstructionBreakpoint:
 	mtspr	SPRN_SPRG_SCRATCH0, r10
-	mtspr	SPRN_SPRG_SCRATCH1, r11
-	lis	r10, (instruction_counter - PAGE_OFFSET)@ha
-	lwz	r11, (instruction_counter - PAGE_OFFSET)@l(r10)
-	addi	r11, r11, -1
-	stw	r11, (instruction_counter - PAGE_OFFSET)@l(r10)
+	lwz	r10, (instruction_counter - PAGE_OFFSET)@l(0)
+	addi	r10, r10, -1
+	stw	r10, (instruction_counter - PAGE_OFFSET)@l(0)
 	lis	r10, 0xffff
 	ori	r10, r10, 0x01
 	mtspr	SPRN_COUNTA, r10
 	mfspr	r10, SPRN_SPRG_SCRATCH0
-	mfspr	r11, SPRN_SPRG_SCRATCH1
 	rfi
 #else
 	EXCEPTION(0x1d00, Trap_1d, unknown_exception, EXC_XFER_EE)
@@ -966,17 +973,3 @@ swapper_pg_dir:
  */
 abatron_pteptrs:
 	.space	8
-
-#ifdef CONFIG_PERF_EVENTS
-	.globl	itlb_miss_counter
-itlb_miss_counter:
-	.space	4
-
-	.globl	dtlb_miss_counter
-dtlb_miss_counter:
-	.space	4
-
-	.globl	instruction_counter
-instruction_counter:
-	.space	4
-#endif
Signature · 1 line
2.13.3
#15 lore raw
In reply to Christophe Leroy

Hi

Please note that I'm currently deeply reworking this serie in order to make it more clear hence more simple to review. So don't consider taking it yet (except maybe the first patch of the serie which is a bugfix).

Christophe

Le 29/05/2018 à 17:50, Christophe Leroy a écrit :

Quoted 88 lines

The purpose of this serie is to implement hardware assistance for TLB table walk on the 8xx.

First part is to make L1 entries and L2 entries independant.
For that, we need to alter ioremap functions in order to handle GUARD attribute at the PGD/PMD level.

Last part is to reuse PTE fragment implemented on PPC64 in order to not waste 16k Pages for page tables as only 4k are used.

Tested successfully on 8xx.

Successfull compilation on following defconfigs (v3):
ppc64_defconfig
ppc64e_defconfig

Successfull compilation on following defconfigs (v2):
ppc64_defconfig
ppc64e_defconfig
pseries_defconfig
pmac32_defconfig
linkstation_defconfig
corenet32_smp_defconfig
ppc40x_defconfig
storcenter_defconfig
ppc44x_defconfig

Changes in v3:
- Fixed an issue in the 09/14 when CONFIG_PIN_TLB_TEXT was not enabled
- Added performance measurement in the 09/14 commit log
- Rebased on latest 'powerpc/merge' tree, which conflicted with 13/14

Changes in v2:
- Removed the 3 first patchs which have been applied already
- Fixed compilation errors reported by Michael
- Squashed the commonalisation of ioremap functions into a single patch
- Fixed the use of pte_fragment
- Added a patch optimising perf counting of TLB misses and instructions

Christophe Leroy (14):
Revert "powerpc/8xx: Use L1 entry APG to handle _PAGE_ACCESSED for
CONFIG_SWAP"
powerpc: move io mapping functions into ioremap.c
powerpc: make ioremap_bot common to PPC32 and PPC64
powerpc: common ioremap functions.
powerpc: use _ALIGN_DOWN macro for VMALLOC_BASE powerpc/nohash32: allow setting GUARDED attribute in the PMD directly powerpc/8xx: set GUARDED attribute in the PMD directly powerpc/8xx: Remove PTE_ATOMIC_UPDATES
powerpc/mm: Use hardware assistance in TLB handlers on the 8xx powerpc/8xx: reunify TLB handler routines
powerpc/8xx: Free up SPRN_SPRG_SCRATCH2
powerpc/mm: Make pte_fragment_alloc() common to PPC32 and PPC64 powerpc/mm: Use pte_fragment_alloc() on 8xx
powerpc/8xx: Move SW perf counters in first 32kb of memory

  arch/powerpc/include/asm/book3s/32/pgtable.h |  29 +-
  arch/powerpc/include/asm/book3s/64/pgtable.h |   2 +
  arch/powerpc/include/asm/highmem.h           |  11 -
  arch/powerpc/include/asm/hugetlb.h           |   4 +-
  arch/powerpc/include/asm/machdep.h           |   2 +-
  arch/powerpc/include/asm/mmu-8xx.h           |  38 +--
  arch/powerpc/include/asm/mmu_context.h       |  53 ++++
  arch/powerpc/include/asm/nohash/32/pgalloc.h |  56 +++-
  arch/powerpc/include/asm/nohash/32/pgtable.h |  77 +++--
  arch/powerpc/include/asm/nohash/32/pte-8xx.h |   6 +-
  arch/powerpc/include/asm/nohash/pgtable.h    |   4 +
  arch/powerpc/include/asm/page.h              |   2 +-
  arch/powerpc/include/asm/pgtable-types.h     |   4 +
  arch/powerpc/kernel/head_8xx.S               | 409 +++++++++++----------------
  arch/powerpc/mm/8xx_mmu.c                    |  12 +-
  arch/powerpc/mm/Makefile                     |   2 +-
  arch/powerpc/mm/dma-noncoherent.c            |   2 +-
  arch/powerpc/mm/dump_linuxpagetables.c       |  32 ++-
  arch/powerpc/mm/hugetlbpage.c                |  12 +
  arch/powerpc/mm/init_32.c                    |   6 +-
  arch/powerpc/mm/{pgtable_64.c => ioremap.c}  | 239 ++++++----------
  arch/powerpc/mm/mem.c                        |  16 +-
  arch/powerpc/mm/mmu_context_book3s64.c       |  44 ---
  arch/powerpc/mm/mmu_context_nohash.c         |   4 +
  arch/powerpc/mm/pgtable-book3s64.c           |  72 -----
  arch/powerpc/mm/pgtable.c                    |  82 ++++++
  arch/powerpc/mm/pgtable_32.c                 | 167 +++--------
  arch/powerpc/mm/pgtable_64.c                 | 177 ------------
  arch/powerpc/platforms/Kconfig.cputype       |  19 ++
  29 files changed, 657 insertions(+), 926 deletions(-)

copy arch/powerpc/mm/{pgtable_64.c => ioremap.c} (53%)